An Unexpected Small Church Problem

Before stepping out into college, my experience in church was extremely small. I'd been a Christian for about two and a half years. Most of that time was spent in a church of about 200 to 250 people in a small town called Canal Fulton, Ohio. The church was a great size — enough people that you could know everybody's faces, but not so many that it would feel weird to walk in as a new person.

Before that, I'd gone to a small Methodist church. This was before I was even a follower of Jesus. We'd go there every now and then, and there were probably 10 kids in the youth group with me. Maybe 40 to 50 people total. But that makes sense — it was in a really small town in Ohio, probably less than 2000 people in the community at the time. My graduating high school class was less than 100. I was used to tiny churches.

Then I moved out on my own. For the first year of college, I went to school in Youngstown, a larger, at the time dying city about an hour and a half from my hometown. For the first time since becoming a Christian, I didn't have a large number of believers surrounding me and helping me learn what it means to follow Jesus. I had one really good friend who knew him. Other than that, I didn't know anyone.

I started looking for churches I could attend while in school. The first one I tried was in the denomination I'd come out of — the Christian and Missionary Alliance. (By the way, if anyone's looking for a church in a denomination, they're pretty darn solid.) The church was large. My memory's a little fuzzy, but I think they were around 2,000 people. This was 2002, so it was a while ago.

I walked in the doors. Carried in my cup of coffee, hiding behind it a little. Gave a nod to the person at the door, turned into the sanctuary. Sat down. Stood and worshipped. Sat and listened. Left.

All without ever talking to another human being.

It was vastly different from what I'd experienced before. So different that I immediately rejected it. Nope. Can't do that. Can't be part of something where I can't know everybody. I tried a couple more times, but it bothered me that I never saw the same faces twice. I didn't like it.

So for a while, my brain went straight to "big is bad."

It's kind of weird, right? The way our initial interactions with something can color how we see it for a very long period of our lives. That experience — what I'd come out of, what I'd stepped into, and the peace I'd felt before I walked in those doors — all compounded to make me think these big churches might not be all that good.

Today

I say that as a person currently working at what is, by American standards, a big church. We're pushing about 700 people at Mosaic now. I'm the executive pastor. I'm one of the faces most people see, but most of them probably don't actually get to talk to me. We're running multiple services, growing steadily, and I have zero doubt we'll be passing the 1,000-person mark, then 2000, at some point. If Jesus wants us that big, we will be. If he doesn't, we'll be whatever size he wants.

I'm not as scared of big churches as I used to be. And there's one specific reason I want to point out.

While big churches can sometimes be bad because they don't see people, small churches have a different problem.

Sometimes, they can be bad because they see people too well.

That might sound like a crazy statement. Hear me out.

The Prayer Ambush

One of my good friends growing up told me about a church he visited. He and his wife were trying to find a church for their family — they had two young daughters at the time. This was probably 30 to 35 years ago.

They stepped into a church that seemed okay at first. The kids were taken to a separate room for children's ministry, which is normal enough. But then, while the pastor was preaching, he stopped. Looked over. And said something. I am paraphrasing a story from a few decades ago, so I’m paraphrasing, but he basically said "Ladies and gentlemen, we have some new guests today. Let's take a moment and go welcome and pray for them."

The entire church got up, walked over, surrounded them, put hands on their shoulders, and started praying for them — simultaneously and loudly

My friend can laugh about it now. At the time, it was extremely disconcerting. People were pressing in harder than he was comfortable with. He didn't feel like he could move. And while this was happening to him and his wife, he had no idea what was happening with his kids in the other room.

I don't know that church's theology. I don't know their intentions. But I know one thing: it was not possible to walk in there and quietly learn who they were without being noticed.

That's an extreme example. But the underlying problem plagues a lot of smaller churches in subtler ways.

The Attention Cliff

I can't tell you how many times I've talked to people about past church hurt, and the hurt is tied to one thing: the amount connection they feel to the church as a whole, and the amount of access they have to people in leadership is less than it used to be.

Church can often feel like a game of "I was in, then I was out. I had access, and then I didn't."

Here's a framework that helps explain why this is so painful. I learned this from the wife of one of the pastors I used to work with years ago (super smart lady, counselor by trade), and it stuck: our contentment in life is largely based on how closely our expectations match reality. High expectations that don't get met lead to discontentment. Low expectations that get exceeded lead to gratitude. Simple math.

Why does that matter here?

In a church of 100 to 200 people, it is very easy for pastors and leaders to notice when someone new walks through the door. And if your goal is to tell people about Jesus and grow the kingdom of God, it's natural to focus time and energy on the faces you don't recognize.

A new guest walks in. Maybe the church won't be as dramatic as the one that surrounded my friend. But they'll get more friendly smiles than they expected. People will introduce themselves in a kind way and ask good questions. The pastor may notice and walk up, offer to grab coffee, ask what they're looking for. They might get plugged into a small group right away. They'll get reached out to, called on, noticed when they're missed.

People see them. And it feels incredible.

Until they're no longer the new face. Until the next new person shows up. Because that level of attention that was pointed at them for weeks or months is now pointed at someone else.

No more weekly coffee meetings with the lead pastor. No more being part of the small group that meets at his house. No more being best friends with the elders. Because those mission-driven people — the ones who genuinely want to tell new people about Jesus — now know you well enough to know that you know Jesus. So their attention moves to the person who might not.

This can be soul-crushing if your expectation was that this is the level of relationship you'd always have. That expectation cannot be reality in a small, mission-driven church. Some churches do this more than others. Some are worse at it than others. The worst ones can probably rightfully be accused of something with a name: love bombing.

What is Love-Bombing?

If you're not familiar with the term, love-bombing is a pattern where, at the very beginning of a relationship, a person or group shows an overwhelming amount of love, attention, kindness, and care — to the point that it can be nearly overpowering. UC Boulder puts it this way: “‘Love-bombing’ is a term used to describe overwhelming and often incongruent affection during the early stages of a friendship or dating relationship. One person goes above and beyond to please the other person by giving excessive compliments, pushing for commitment or more time together, making grand gestures and sending over-the-top gifts.”1 The purpose is to give the other person a taste of what a deep relationship could feel like. It creates a false impression of what life in this relationship will always look like.

Then the intensity drops. The person on the receiving end isn't processing reality clearly anymore. They're chasing what they used to have, desperately clinging to the early days instead of seeing the current state of the relationship with clear eyes.

Love-bombing is notorious in cults. It's notorious in abusive relationships. "He swept me off my feet, but I didn't understand he was an abuser" is a painfully common refrain. "They just seemed so kind and genuine. How could I have known?"

Church Plants are notorious for love-bombing people on accident.

The leadership is so focused on reaching people that they pour hours and hours into every new person's life each week — without realizing that those new people, if they successfully become established members, will eventually experience a sharp drop in attention. And there's a saying in growing churches: people who come to your church and stick often think the perfect size for a church is exactly as large as it was when they arrived. That tends to be because they equate that moment with the level of attention they were comfortable receiving. If the church is growing, it will not be able to maintain that level.

Where mid-to-large churches can actually excel

This is where I think churches in the mid-to-large range have a real chance of getting it right.

We can build healthy expectations for people who walk through our doors about what a relationship with Jesus — in the context of being part of this particular body — actually looks like. We're small enough that people will see familiar faces in the crowd each week. They can be noticed by the person sitting next to them, or by people they've started serving alongside in a volunteer team or connecting with in a small group. They can be part of a smaller community that gives them the social connection they're looking for, while simultaneously having a relationship with the larger body of the church.

When connection is done well, people walk in with good expectations for when they'll be seen, how they'll be seen, and who will know what's going on in their lives. They understand what role the church can play in their story. They won't feel ignored. They won't be dropped like they don't matter. But they also won't receive hyper-individualized attention from every staff member they encounter.

They'll form healthy relationships. Mutual friendships that help each person grow closer to Jesus as they walk together.

And nobody has to get ambush-prayed in a sanctuary to make that happen.

AI weekly roundup: April 1-7, 2026

In case you missed it — the biggest stories from the AI world this week, deduplicated from the daily digests.

Models and capabilities

  • Anthropic's "Claude Mythos" leaked. A CMS misconfiguration exposed ~3,000 unpublished Anthropic assets, revealing a next-gen model called Mythos (codename Capybara) that Anthropic describes as "a step change" beyond Opus 4.6. Internal benchmarks show dramatic improvements in coding, reasoning, and cybersecurity. Anthropic is privately warning governments that Mythos's cyber capabilities "far outpace defenders." No public release date — Q2 2026 at earliest. Fortune · CNN

  • Google released Gemma 4 under Apache 2.0. Four sizes (2B, 4B, 26B MoE, 31B Dense), 256K context, native vision and audio, 140+ languages. The 31B flagship scores 89.2% on AIME 2026 math (up from 20.8% in Gemma 3) and 86.4% on agentic tool use (up from 6.6%). Runs on consumer hardware down to mobile. Google Blog · The Register

  • OpenAI retired GPT-4o; GPT-5.4 gets native computer use. GPT-4o is fully gone. GPT-5.4 can now operate computers and execute cross-application workflows — putting OpenAI on par with Claude's computer use capabilities. AI and News

  • DeepSeek V4 reportedly weeks away. 1-trillion-parameter MoE model (~37B active), 1M-token context, native multimodal, trained entirely on Huawei Ascend chips (no Nvidia). Apache 2.0 license, estimated $5.2M training cost. NxCode

  • Open models have crossed the frontier threshold. LangChain analysis confirms open models now rival proprietary ones for core agent tasks, with cost and latency advantages. LangChain Blog

Industry and funding

  • OpenAI closed a record $122B round at $852B valuation. Amazon ($50B), SoftBank ($30B), Nvidia ($30B), and for the first time ~$3B from retail investors. Generating $2B/month in revenue with 900M+ weekly active users. IPO expected later this year. Bloomberg · TechCrunch

  • Q1 2026 startup funding hit $297B — a 2.5x increase over Q4 2025, driven by the OpenAI and Anthropic ($30B at $380B valuation) mega-rounds. TechCrunch

  • ChatGPT now has conversational ads. OpenAI partnered with Smartly to bring interactive ad units inside ChatGPT — ads you can chat with. Pilot has crossed $100M annualized revenue, 600+ advertisers. The Next Web

  • Anthropic acquired Coefficient Bio for ~$400M, forming a new healthcare/life sciences group. Signals expansion beyond pure-play AI. TechCrunch

  • Anthropic formed a PAC for influencing AI policy and regulation. TechCrunch

  • Oracle cutting 20K-30K jobs to redirect $8-10B toward AI infrastructure. Block (Square) also eliminated 4,000 roles, with Dorsey explicitly citing AI redundancy. The Neuron

Tools and products

  • Cursor 3 shipped — agent-first IDE. Ground-up redesign around fleets of autonomous agents. Multi-repo, local-cloud handoff, MCP plugin marketplace, and a sidebar that aggregates agents from desktop, mobile, Slack, GitHub, and Linear. Cursor Blog

  • Claude Code hit #1 on GitHub trending with 104K stars (10,749 added in a single day). Four versions shipped in the first week of April alone (v2.1.89–2.1.92): 60% faster large-file writes, per-model cost breakdowns via /cost, 500K character MCP result persistence, and Bedrock setup wizard. Releasebot

  • OpenAI Codex went pay-as-you-go for teams. No fixed seat fee — pure token billing. ChatGPT Business dropped from $25 to $20/seat. New Codex members get $100 in credits (up to $500/team). OpenAI

  • Microsoft released three MAI foundation models. MAI-Transcribe-1 (25 languages, 50% lower GPU cost), MAI-Voice-1 (60 seconds of audio in under 1 second), MAI-Image-2 (#3 on Arena.ai). First proprietary multimodal models from Microsoft's superintelligence team. Microsoft

  • Microsoft open-sourced VibeVoice — 1.5B-parameter voice AI covering TTS and ASR, 50+ languages, 60-minute long-form audio in a single pass. Best open-source voice option available. GitHub

  • OpenClaw exploded to 210K+ GitHub stars — open-source personal AI assistant running locally, connecting to 50+ integrations (WhatsApp, Telegram, Slack, Discord, Signal, iMessage). Fastest-growing open-source project in GitHub history. GitHub

  • Anthropic started billing third-party Claude tools separately as of April 4. First time tool pricing has been unbundled from base model access. NewsBytesApp

MCP ecosystem

  • Pinterest published the most detailed enterprise MCP case study yet. Domain-specific cloud-hosted servers, central registry, two-layer auth (user JWTs + mesh identities), mandatory security/legal review. 66,000 invocations/month, 844 active users, 7,000 hours saved/month. InfoQ

  • Google announced fully-managed MCP servers for BigQuery, Maps, Cloud Run, Cloud Storage, AlloyDB, Cloud SQL, Spanner, Looker, Pub/Sub, and Dataplex. Google Cloud

  • MCP 2026 roadmap published. Four priorities: transport scalability (Streamable HTTP), agent-to-agent communication (Q3), governance maturation, and enterprise readiness (OAuth 2.1 in Q2). 97 million monthly SDK downloads, 5,800+ community servers. MCP Blog

  • MCP Go SDK security fix (CVE-2026-34742) — DNS rebinding protection now enabled by default. Update if running Go-based MCP servers. GitLab Advisory

  • MCP tool poisoning attack disclosed. Invariant Labs showed that malicious instructions embedded in MCP server documentation can influence AI behavior without users knowing. Vet your servers. The New Stack

  • Smithery.ai now indexes 7,300+ MCP servers. Smithery.ai

Security

  • LiteLLM supply chain attack compromised 1,000+ AI environments. A malicious PyPI package injected into the popular LLM routing library. AI recruiting unicorn Mercor ($10B valuation) confirmed as a victim, with Lapsus$ claiming 4TB of stolen data. If you use LiteLLM, update immediately and rotate all API keys. TechCrunch· The Record

Research

  • AI Scientist-v2 passed peer review. Sakana AI's autonomous research agent proposed hypotheses, ran experiments, and wrote a paper accepted at an ICLR workshop (score 6.33 vs. 6.0 threshold). First fully AI-generated paper to pass peer review. arXiv

  • Grok 4.2 uses multi-agent debate to cut hallucination by 65%. Four specialized agents debate before producing an answer, dropping hallucination from ~12% to ~4.2% at only 1.5-2.5x compute cost. Medium

Deadlines

  • April 19: Claude Haiku 3 deprecated — migrate to Haiku 4.5

  • April 22: Anthropic Enterprise Agents virtual briefing

  • April 30: 1M token context window beta retiring for Sonnet 4.5 and Sonnet 4 — requests over 200K tokens will error after this date

Compiled from AI Daily Digests, April 1-7, 2026 - VIA Claude Cowork

Keep reading